NeurIPS 2021 Reveals Outstanding Paper and Test of Time Award Winners
NeurIPS 2021 officially announces the recipients of its Outstanding Paper Awards, the Test of Time Award, and the new Datasets and Benchmarks Track Best Paper Awards ahead of next week's conference.
NeurIPS 2021 officially announces the recipients of its prestigious Outstanding Paper Awards, the Test of Time Award, and the newly introduced Datasets and Benchmarks Track Best Paper Awards. With the conference set to begin next week, the program chairs express deep gratitude to the dedicated community members who lead the award selection process and provide essential subject-matter expertise.
A committee of prominent researchers selects six papers to receive the Outstanding Paper Award based on their excellent clarity, insight, creativity, and potential for lasting impact. Among the highlighted winners is "A Universal Law of Robustness via Isoperimetry" by Sébastien Bubeck and Mark Sellke, which presents a theoretical model explaining why state-of-the-art deep networks require significantly more parameters than necessary to smoothly fit training data.
This specific award-winning paper demonstrates that the number of parameters needed for a function to smoothly interpolate training data scales as nd, where n represents the number of training examples and d is the data dimensionality. This simple and elegant theory contrasts with conventional beliefs and aligns closely with empirical observations regarding the actual size of robust models in modern deep learning applications.